The $71 billion hedge fund has been using an equity chatbot over the past year, the firm’s CTO, Umesh Subramanian, revealed this week at a conference in New York.

The Citadel AI Assistant scans transcripts and filings, summarizes brokerage research, and flags potential risks for the firm’s equities team, delivering the material much faster than a human researcher. The tool is now part of the daily workflow for most of the firm’s equities investors.

Subramanian said the firm still relies on human judgment, according to Reuters.

“We are also careful that it’s not used in the wrong way,” he said Wednesday. “We don’t want PMs offloading their human investment judgment to AI. This is a tool to further accelerate their research process.”

Ken Griffin, Citadel’s billionaire founder, said in October that while the technology can boost efficiency, it’s unlikely to produce market-beating returns. Subramanian echoed that point at the conference:

Other hedge funds have been more bullish on the technology. In March, Bridgewater’s CEO said that technology has generated “unique alpha that is uncorrelated to what our humans do.”

Other hedge fund giants have disclosed at least some aspects of how they’re adopting the technology. I’m unaware of any firm that has publicly said they’re not embracing the technology in some capacity. Here’s a rundown of some of AI Street’s coverage:

This list is not exhaustive. Let me know if there are any firms that I missed by replying to this email.

Separately, I’m doing some reporting on hedge funds building out GraphRAG for a more efficient data and research architecture as well as using hypothetical document embeddings (HyDE), which I’m trying to wrap my head around. If you’re building in this area, please email me. ([email protected])

While hedge fund managers have mixed views on whether AI can generate alpha, the technology is becoming standard infrastructure across the industry. The question is no longer whether firms will use these tools, but how deeply they’ll rely on them.

Here’s a round up of some of this year’s most popular AI and finance research:

AI is trained with data. It’s not given rules to follow.

So instead of deciding what may be the most important factor or independent variable, researchers let the data decide.

That’s the idea behind Artificial Intelligence Asset Pricing Models.

Semyon Malamud, one of the authors, explained the intuition: you can’t understand a stock in isolation. Its meaning comes from its context relative to other stocks.

Key findings:

Instead of relying on one frontier model, BlackRock built three AI “agents” that mimic different analyst roles:

Each agent analyzes a stock independently, then enters a round-robin debate. Disagreements are argued until the agents reach consensus on whether to BUY or SELL — a process designed to mimic an investment committee.

The system runs on Microsoft's AutoGen framework using GPT-4o, with custom tools for each agent: document parsing for 10-Ks, news summarization, and volatility calculators.

Two papers from the same research group explore what happens when you train transformer models, the architecture behind ChatGPT, on bond and equity portfolio data.

Their findings suggest that the models excel not only at predicting the next word in a sentence but also at anticipating bond downgrades and spotting patterns that traditional financial metrics overlook. Check out these stories that I wrote for the Chicago Booth Review.

IBM researchers showed they can force an LLM to produce the same answer every time by using a smaller, deterministic model.

LLMs are probabilistic, which makes consistency a challenge. IBM tested whether they could remove that variability.

What they did:

What they found:

Major technology waves rarely take off until the underlying plumbing becomes standardized. The internet needed TCP/IP. The web needed HTML. Databases needed SQL.

AI lacks this foundation. The technology is essentially three years old, and organizations are still scrambling to catch up. Determining technical protocols takes time, but we’re starting to see these organizational standards emerge.

Hedge funds are already treating AI talent as essential infrastructure, with Marshall Wace now passing tens of millions of dollars in technology and quant hiring costs directly to clients, according to Bloomberg. Because the fund relies heavily on AI and LLMs, they argue that hiring and retaining elite risk employees is "vital" to maintaining a competitive edge—and therefore a justifiable expense for clients.

As AI becomes more entrenched in financial services, regulators are evolving their oversight frameworks to keep pace.

Over the last few years, we’ve seen the emergence of Chief AI Officer to spearhead all of the different use cases. And management professors are recommending them. (Why Your Company Needs a Chief Data, Analytics, and AI Officer). Just this week we saw a few hires in this space:

AI is becoming standardized, but the organizational infrastructure required to scale it is still being built and has a long way to go. This foundation is coming, but it will take time.

Advertise on AI Street to reach a highly engaged audience of decision-makers at firms including JPMorgan, Citadel, BlackRock, Skadden, McKinsey, and more. Sponsorships are reserved for companies in AI, markets, and finance. Email me ([email protected]) for more details.